Kalman filtering for estimation of closed-die forging load based on shop floor data

Abstract

Estimation of forging load in a closed-die forging process is important for process planners and designers. Physics-based models of the process require very high computational time as well as accurate input data. This work estimates the forging load by extracting the information from similar products in the shop floor. Accuracy of the estimate is optimized by means of a Kalman filter. A novel feature of this work is that instead of one deterministic estimate, three estimates, viz. lower, upper and most likely, are obtained. For demonstrating the efficacy of the proposed methodology, finite element method simulations using ABAQUS are used in lieu of real shop floor data. Six different products each having eight models are considered. Forging is supposed to be carried out with as well as without lubrication. In different cases, Kalman filtered most-likely estimate came very close to actual (FEM simulated) forging load and in no case deviation was more than 9%. The estimation error keeps on reducing with availability of data in an exponential manner. In an ideal case of no fluctuation in the measured actual (FEM simulated) forging load, error reduced from 30 to 5 kN after 7 data; further 13 data reduced the error to 1.7 kN. The interval of estimation (i.e. the difference between upper and lower estimates) also keeps on reducing with the availability of more data. For example, for an axisymmetric product, availability of 6 more data reduced the range of estimation from 62 to 25 kN. This establishes the efficacy of Kalman filter. In the proposed procedure the data is stored in an open source relational database management system, the MySQL, which can be retrieved easily. In Industry 4.0, shop floor data is easily available. Hence, the proposed method can be applied readily in real production shops.